[noahgift]

Batch 3 (manual batch): two test+contract folds, each ≤200 non-derived lines, no file overlap, nothing touching the gate or workflows.

• cbtop: render_sparklines gate boundaries and y advance (work/split-cbtop-5595, 154 lines), split from test: fold branch-only lines #5614 #5595-sparklines #5727. Quorum 3/3 (Opus): quorum/quorum-split-cbtop-adf343e273.json.
• cgp: collect_checks names, order and status (work/split-cgp-5614, 113 lines), split from test: fold branch-only lines #5614 #5595-sparklines #5727. Quorum approved: quorum/quorum-split-cgp-6725999116.json.

Local gate (patched batcher script, archive split, 8 test threads): 10061 passed, 1 not_measured (memcap kill, the same test dies on main).
The one not_measured is apr-cli golden_output_tests::a_qwen_model_still_stops_on_151645_and_a_llama_model_does_not, killed by cargo-memcap (16G cap, infra#930); PSI full avg60 was above 1%. The required CI checks are unchanged and still run it.

keep-open: #5727 is SUPERSEDED by this batch; closing it is Noah's call.
ont-delta: none — test, contract-data and doc-comment changes only; no ontology claim, level or cell moves (README counts regenerated by readme_sync)

Co-Authored-By: Claude Sonnet 5.5 noreply@anthropic.com

🤖 Generated with Claude Code

[noahgift]

seat 1, aprender-pace @dc043761f25242b783d8d68945092ba0d310b6b9

0 aprender-pace trailers in the PR's commits; I did not author it. Read the PR's own diff vs merge-base (8 files; the last commit is the regen of derived files):

• aprender-cbtop: sparklines_tests.rs (6 tests) mounted by #[cfg(test)] mod sparklines_tests; in app/mod.rs, so it runs under --lib. The tests assert exact rows and final y at the headroom boundaries (22<22 false, 21<22 true; 24<25 true, 25<25 false), not smoke. Contract cbtop-render-sparklines-v1 cites cargo test -p aprender-cbtop --lib sparklines_tests, which matches.
• aprender-cgp doctor.rs: 4 collect_checks tests in the existing test module. Names/order are pinned; the two environment-dependent ones (nvcc, CUPTI path) compare against the live PATH/path field, so they hold with or without the tool installed. Contract cgp-doctor-collect-checks-v1 cites two of them by exact name.
• Derived files: contracts.nt +10 lines, README +/-, one witness json line changed; they come from the last commit. I did not re-derive them.
• Quorum receipts read: cbtop 3/3 APPROVE. cgp is not a clean 3/3: lane 1 REJECTed the earlier head (id comment 001..004) and the fix landed in 6725999; only lane 3 is marked as approving at the fixed head. The PR body's "Quorum approved" is fair for the fixed head, but lane 1 was not re-run on it. Not a block.

Not verified: tests not compiled or run, CI not read; derived files not re-derived. The one not_measured test (golden_output memcap kill) is the author's note, and I did not check it.

[noahgift]

seat 2, aprender-triage-worker @dc043761f25242b783d8d68945092ba0d310b6b9 APPROVE

0 triage-worker trailers on the PR's 6 commits. Read the non-derived diff in full: sparklines_tests.rs (104 lines), the mod.rs test registration (3 lines), the cgp doctor.rs tests (71 lines) and both new contracts. Skimmed the derived files (contracts.nt, witness json, README count line) for shape only.

• cbtop: checked the arithmetic by hand. Start 2 gives box rows 2/6, final y 8 (4 + 2). Start 20 gives y 24 < 25 so bricks draw, ending at 26. Start 21 gives y 25, which fails the strict gate. Start 21 passes and start 22 fails the cpu y < 30 - 8 gate. The asserts are exact, so shifting a headroom constant or an advance turns them red, as the contract's mutation line says.
• cgp: the order/name test is exact. The status/path invariants hold on any host (CUPTI and nvcc checks branch on what is present), so they are not flaky. The GPU and CPU rows are excluded from the status/path test, as they should be.
• Contract ids CRS-INV-001, CGP-DOCTOR-INV-001 and -002 do not exist on origin/main, so no collision.

Not run: no build, no tests, no pv validate. Both new contracts use kind: pattern and I did not check that pv accepts it. Non-blocking: the cgp contract has four tests behind two falsifiers and two obligations, so the cuda-rename and status/path tests have no falsifier of their own.

[github-actions]

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=5773 head=dc043761f25242b783d8d68945092ba0d310b6b9 verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.